Upload sales_inference.py with huggingface_hub

sales_inference.py

@@ -0,0 +1,91 @@
+# sales_inference.py - Simple Azure OpenAI Only Implementation
+import os
+import numpy as np
+import torch
+from openai import AzureOpenAI
+from stable_baselines3 import PPO
+from stable_baselines3.common.torch_layers import BaseFeaturesExtractor
+import torch.nn as nn
+from dataclasses import dataclass
+from typing import List, Dict
+
+# Azure OpenAI Configuration
+openai_client = AzureOpenAI(
+    api_key=os.getenv("AZURE_OPENAI_API_KEY"),
+    api_version="2023-05-15",
+    azure_endpoint=os.getenv("AZURE_OPENAI_ENDPOINT")
+)
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+class CustomCNN(BaseFeaturesExtractor):
+    def __init__(self, observation_space, features_dim: int = 64):
+        super().__init__(observation_space, features_dim)
+        n_input = observation_space.shape[0]
+        self.cnn = nn.Sequential(
+            nn.Linear(n_input, 512), nn.ReLU(),
+            nn.Linear(512, 256), nn.ReLU(),
+            nn.Linear(256, features_dim), nn.ReLU(),
+        ).to(device)
+    def forward(self, obs): return self.cnn(obs)
+
+@dataclass
+class SalesAgent:
+    model_path: str
+    
+    def __init__(self, model_path):
+        self.model = PPO.load(model_path, device=device, custom_objects={
+            "policy_kwargs": dict(
+                activation_fn=nn.ReLU,
+                net_arch=[dict(pi=[128, 64], vf=[128, 64])],
+                features_extractor_class=CustomCNN,
+                features_extractor_kwargs=dict(features_dim=64)
+            )
+        })
+    
+    def get_embedding(self, text):
+        response = openai_client.embeddings.create(
+            model=os.getenv("AZURE_EMBEDDING_DEPLOYMENT", "text-embedding-3-large"),
+            input=text
+        )
+        emb = np.array(response.data[0].embedding, dtype=np.float32)
+        # Adjust to expected dimension 3072
+        return np.concatenate([emb[:3072], np.zeros(max(0, 3072-len(emb)))]) if len(emb) < 3072 else emb[:3072]
+    
+    def predict_conversion(self, conversation_text, turn_number):
+        emb = self.get_embedding(conversation_text)
+        metrics = np.array([0.5, 0.5, turn_number, 0.5, min(1.0, turn_number/20)], dtype=np.float32)
+        state = np.concatenate([emb, metrics, [turn_number], np.zeros(10)])
+        with torch.no_grad():
+            action, _ = self.model.predict(state, deterministic=True)
+        return max(0.0, min(1.0, float(action[0])))
+    
+    def analyze_conversation(self, conversation):
+        results = []
+        text = ""
+        for i, turn in enumerate(conversation):
+            text += f"{turn['role']}: {turn['content']}\n"
+            if turn['role'] == 'assistant':
+                prob = self.predict_conversion(text, i//2+1)
+                results.append({'turn_number': i//2+1, 'response': turn['content'], 'probability': prob})
+        final_prob = results[-1]['probability'] if results else 0.0
+        return {'conversation_results': results, 'final_probability': final_prob, 
+                'summary': f"Predicted final conversion: {final_prob*100:.1f}%"}
+
+# Usage example
+if __name__ == "__main__":
+    # Download model from HuggingFace
+    from huggingface_hub import hf_hub_download
+    model_path = hf_hub_download(repo_id='DeepMostInnovations/sales-conversion-model-reinf-learning', filename="sales_model.zip")
+    
+    # Initialize agent
+    agent = SalesAgent(model_path)
+    
+    # Test conversation
+    conv = [
+        {"role": "user", "content": "I'm interested in your product"},
+        {"role": "assistant", "content": "Great! What's your team size?"}
+    ]
+    
+    result = agent.analyze_conversation(conv)
+    print(result['summary'])